Malicious ChatGPT Links Exploit User Sessions to Access Private Chats

• •A vulnerability in ChatGPT allowed malicious links to run JavaScript in user sessions.
• •Attackers could access private chats and enumerate conversation lists without user consent.
• •OpenAI fixed the issue within eight days of disclosure, classifying it as a P1 vulnerability.

Researchers from Parameter discovered a vulnerability in ChatGPT that allowed malicious links to execute attacker-controlled JavaScript within the signed-in session of users. This flaw enabled attackers to access private conversations and send data to external servers. The vulnerability was reported to OpenAI on October 18, 2026, and was classified as P1, the highest severity level. OpenAI released a fix on October 26, 2026. The attack exploited the HTML Preview feature and a graph rendering process, allowing unauthorized access to user data without any additional actions required from the victim. No real-world exploitation has been reported, and the attack was demonstrated in a controlled environment. The issue was linked to how ChatGPT handled cross-frame messaging and HTML content. Parameter emphasized that the flaw could have allowed full control over a victim's ChatGPT account.